Understand data flow and AI access
See how prompts, model context, queries, files, previews, credentials, and permissions participate in the workflow.
Unity connects your browser, its backend, your data sources, and an AI service. The assistant uses authorized context and tool results to answer questions and build report content.
This guide explains the product's data flow. It does not replace deployment-specific policies or promise that every result remains only in your database.
Follow a request
- Your browser submits a message and report context to Unity.
- Unity authenticates the account and checks report/resource permissions.
- The assistant receives relevant model, report, conversation, and instruction context.
- Tools retrieve additional details, execute queries, inspect files, or save changes as needed.
- Selected results and diagnostics return to the AI workflow.
- Unity saves the conversation and supported report changes and returns them to the browser.
For HTML work, the loop can also render a page, capture its appearance, and provide that preview for visual inspection.
Information used by the workflow
| Information | Purpose |
|---|---|
| Prompts and conversation history | Understand requests and follow-ups. |
| Entity, field, relationship, and definition metadata | Construct valid model-aware queries and explain calculations. |
| Query text and selected results | Analyze data and validate output. |
| Page and visual definitions | Inspect and edit the existing report. |
| Report instructions | Apply shared analytical and presentation conventions. |
| Relevant reference-file excerpts and previews | Use templates, supporting documents, and original assets. |
| Browser screenshots and diagnostics | Check rendered HTML and repair issues. |
The assistant can retrieve details on demand. It does not automatically receive every row in the database or every attached file in full, but returned query data and relevant file content can enter the AI workflow.
What Unity stores
Reports retain their pages, source/query definitions, visual properties, filters, instructions, reference files, and access information. Conversations retain messages and recorded activity. Supported operations retain history or snapshots used for recovery.
Reference files and reusable report images use private report storage. Imported CSV datasets use a separate data-import storage path. They are distinct resources even when both began as a local file.
Visual results can be cached, and historical chat output can retain data from an earlier query. See Live data for freshness and row limits.
This catalogue does not state a universal retention period or no-storage guarantee.
Source credentials and query identity
Unity stores source connection configuration server-side and uses its credential-handling mechanisms for secrets. Report collaborators use the authorized connection path; ordinary report sharing does not disclose stored credentials.
A report collaborator is not automatically a separate database user. If the connection uses a shared service identity, source permissions follow that identity and Unity's application checks.
Report filters and invisible pages do not by themselves establish per-user source row security.
Report files and original assets
Files attached in chat become report resources, available to authorized collaborators and conversations in that report. A private conversation can therefore refer to a file that is shared at report scope.
The assistant can read excerpts, search sections, preview supported formats, and import original images into a reusable asset library.
File-inspection tools do not execute uploaded macros or scripts or recalculate spreadsheets. An HTML reference preview is static. Authored HTML report visuals use the separate supported browser runtime.
See Reference files.
HTML execution boundary
HTML pages run in a sandboxed iframe. Their browser code receives registered dataset results and supported renderer helpers.
The page cannot use external network fetches, CDN imports, parent-page access, cookies, or browser storage. Unity owns query execution, permissions, lineage, paging, and filter propagation.
Literal values can be authored in HTML, but only registered query results should be described as live data. Source validation and browser previews serve different purposes.
Voice input
Voice input requires browser microphone access. Depending on the available speech path, speech is recognized by browser capabilities or recorded for transcription before text is placed into the draft.
Review the transcript before submitting it. Dictating text is separate from authorizing or executing a report change.
Organizations and access
Organization membership, management authority, team report roles, and direct report grants are separate. Current effective permissions apply to report reads, queries, assistant work, and writes.
Invisible pages are excluded from viewer content and relevant AI/history access paths. An API key runs as its creating account with its scopes and current resource permissions.
An organization Owner can explicitly recover a known company report, but the role does not automatically expose everyone's report content.
Evaluate the flow for your deployment
Review the source identity, exposed model fields, report recipients, relevant files, and actual AI-service configuration for your environment.
Use Query inspection to trace a calculation, Sharing and access to inspect authorization, and the API quickstart to understand personal API keys.
Questions about your environment? Contact us.